Bug#890999: VirtualBox modules built against headers version 4.14.17-1 cannot be loaded by kernel version 4.14.13-1: Unknown symbol __x86_indirect_thunk_r15 (err 0)

2018-02-21 Thread Evgeny Kapun

Package: linux-headers-4.14.0-3-amd64
Version: 4.14.17-1

When I build VirtualBox kernel modules using linux-headers-4.14.0-3-amd64 
version 4.14.17-1, the resulting modules can't be loaded using 
linux-image-4.14.0-3-amd64 version 4.14.13-1. When I build the same modules 
using headers version 4.14.13-1, they load successfully.

Here are some error messages:

systemd[1]: Starting LSB: VirtualBox Linux kernel module...
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_r15 (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_rax (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_rdx (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_r14 (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_rbx (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_r13 (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_r10 (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_rcx (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_rsi (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_r9 (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_r12 (err 0)
kernel: vboxdrv: Unknown symbol __x86_indirect_thunk_r8 (err 0)
virtualbox[6834]: Loading VirtualBox kernel modules...modprobe vboxdrv failed. 
Please use 'dmesg' to find out why ... failed!
virtualbox[6834]:  failed!



Bug#828705: Linux kernel update creates redundant symlinks

2016-06-27 Thread Evgeny Kapun

On 27.06.2016 15:45, Ben Hutchings wrote:

Control: tag -1 wontfix

On Mon, 2016-06-27 at 15:20 +0300, Evgeny Kapun wrote:

On 27.06.2016 15:03, Ben Hutchings wrote:

Doesn't lilo fail if a file specified in its configuration is
missing?

Ben.


LILO has an "optional" option which makes it skip an entry if either
the kernel or the initrd file is missing (or is a broken link).


That's a good point.

I checked whether liloconfig would put the 'optional' keyword in
/etc/lilo.conf, but the config file it generated didn't mention the
symlinks at all.  So I don't think it makes sense to assume that people
use the 'optional' keyword.

There are other boot loaders that I think will use the symlinks during
package updates: elilo, palo and zipl.  They all have similar
configuration syntax to lilo, but it appears that elilo and palo do not
support the 'optional' keyword.

Although the 'old' symlinks will sometimes be redundant, I don't think
this issue outweighs the inconvenience of a failed boot loader update.

Ben.



On the other hand, the current behavior (when both pairs of symlinks are always 
kept) was introduced not so long ago, so most existing setups should work with 
the old behavior. And I can't find anyone reporting the old behavior as a bug, 
so I think that the old behavior isn't actually any more problematic than the 
current one. Maybe add a config option to choose between the old and the 
current behavior?

Also, if keeping all these symlinks is so important, then maintainer scripts 
should always prevent the only remaining installed kernel version from being 
removed, since there is no way to have valid symlinks after that. Now, such 
removal would proceed without any confirmation (unless that version happens to 
be the same as the one that is currently running), and all symlinks will be 
removed.



Bug#828705: Linux kernel update creates redundant symlinks

2016-06-27 Thread Evgeny Kapun

On 27.06.2016 15:03, Ben Hutchings wrote:

Doesn't lilo fail if a file specified in its configuration is missing?

Ben.


LILO has an "optional" option which makes it skip an entry if either the kernel 
or the initrd file is missing (or is a broken link).



Bug#828705: Linux kernel update creates redundant symlinks

2016-06-27 Thread Evgeny Kapun

On 27.06.2016 11:51, Ben Hutchings wrote:

Control: reassign -1 linux-base 4.3
Control: tag -1 moreinfo

On Mon, 2016-06-27 at 03:59 +0300, Evgeny Kapun wrote:

Package: linux-image-4.6.0-1-amd64
Version: 4.6.2-2

When updating the package linux-image-4.6.0-1-amd64 from version
4.6.2-1 to version 4.6.2-2, post-install script printed this:

I: /vmlinuz.old is now a symlink to boot/vmlinuz-4.6.0-1-amd64
I: /initrd.img.old is now a symlink to boot/initrd.img-4.6.0-1-amd64

After that, both /{vmlinuz,initrd.img} and /{vmlinuz,initrd.img}.old
point to the same kernel version, 4.6.0-1-amd64. Aren't these .old
symlinks supposed to point to the previous version?


If only one version is installed, both pairs of links will be pointed
to it.  Previously, one or both pairs could be left broken.

Is there another version installed?

Ben.



No, there is only one version installed. Previously, there was only one pair of 
links in this case (the .old links were removed). I think that the old behavior 
is better because with the current behavior, bootloaders which use these 
symlinks (such as LILO) will create two menu entries for the same kernel 
version, which is confusing. When there is only one pair, such bootloaders will 
only create one entry.



Bug#828705: Linux kernel update creates redundant symlinks

2016-06-26 Thread Evgeny Kapun

Package: linux-image-4.6.0-1-amd64
Version: 4.6.2-2

When updating the package linux-image-4.6.0-1-amd64 from version 4.6.2-1 to 
version 4.6.2-2, post-install script printed this:

I: /vmlinuz.old is now a symlink to boot/vmlinuz-4.6.0-1-amd64
I: /initrd.img.old is now a symlink to boot/initrd.img-4.6.0-1-amd64

After that, both /{vmlinuz,initrd.img} and /{vmlinuz,initrd.img}.old point to 
the same kernel version, 4.6.0-1-amd64. Aren't these .old symlinks supposed to 
point to the previous version?



Bug#812120: update-initramfs fails if busybox is not installed

2016-01-20 Thread Evgeny Kapun

Package: initramfs-tools-core
Version: 0.121

If I try to run update-initramfs while busybox is not installed, I get an error 
message:

E: busybox is required but not installed
E: /usr/share/initramfs-tools/hooks/busybox failed with return 1.

This happens irrespective of the BUSYBOX setting in 
/etc/initramfs-tools/initramfs.conf.



Bug#812120: update-initramfs fails if busybox is not installed

2016-01-20 Thread Evgeny Kapun

On 20.01.2016 22:41, Ben Hutchings wrote:

On Wed, 2016-01-20 at 22:21 +0300, Evgeny Kapun wrote:

Package: initramfs-tools-core
Version: 0.121

If I try to run update-initramfs while busybox is not installed, I get an error 
message:

  E: busybox is required but not installed
  E: /usr/share/initramfs-tools/hooks/busybox failed with return 1.

This happens irrespective of the BUSYBOX setting in 
/etc/initramfs-tools/initramfs.conf.


You have another package installed (maybe cryptsetup) that requires
busybox to work in the initramfs.

Try this:

 grep -r BUSYBOX /usr/share/initramfs-tools/conf-hooks.d

Ben.


Yes, I have cryptsetup installed, but this problem appeared only after 
upgrading initramfs-tools to version 0.121.